Residual Spatiotemporal Autoencoder with Skip Connected and Memory Guided Network for Detecting Video Anomalies

Real-time video anomaly detection and localization still prevail as a challenging task. Autoencoders are expected to give high reconstruction error for abnormal events than normal events while trained on video segments of normal events. Nevertheless, this assumption is not always true in practice. S...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Neural processing letters Jg. 53; H. 6; S. 4677 - 4692
Hauptverfasser: Chandrakala, S., Srinivas, V., Deepak, K.
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Springer US 01.12.2021
Springer Nature B.V
Schlagworte:
Anomalies
Artificial Intelligence
Classification
Complex Systems
Computational Intelligence
Computer Science
Deep learning
Localization
Neural networks
Performance degradation
Video anomaly detection
Residual blocks
Normality modeling
Memory guided network
Spatio temporal autoencoders
ISSN:1370-4621, 1573-773X
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Real-time video anomaly detection and localization still prevail as a challenging task. Autoencoders are expected to give high reconstruction error for abnormal events than normal events while trained on video segments of normal events. Nevertheless, this assumption is not always true in practice. Sometimes the autoencoder offers better generalization. Therefore, it also reconstructs abnormal events well, leading to slightly degraded performance for anomaly detection. To alleviate this issue, we propose a Skip connected and Memory Guided Network (SMGNet) for video anomaly detection. The memory guided network with skip connection help in avoiding loss of meaningful information such as foreground patterns, in addition to memorizing significant normality patterns. The effect of augmenting memory guided network with skip connection in the residual spatiotemporal autoencoder (R-STAE) architecture is evaluated. The proposed technique achieved improved results over three benchmark datasets.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1370-4621
1573-773X
DOI:10.1007/s11063-021-10618-3